.

Geo Informatics 2-2008
article figure 5: full resolu tion of mosaic file with cutline in red color. lution and panchromatic images at a high spatial resolution, it is possible to perform pan-sharpening for both of these images. however, most of the existing techniques, which perform suitably well with medium-resolution images, can hardly satisfy the pan-sharpening of multispectral and panchromatic high resolution images. based on the thorough study and analysis of existing pan-sharpening algorithms and their fusion effects, a new automatic pan-sharpening algorithm has been developed by dr. yun zhang at the university of new brunswick in fredericton, new brunswick, canada. this new technique solved the two major problems in pan-sharpening color distortion and operator dependency. a method based on least squares was employed for a best approximation of the grey level value relationship between the original multispectral, panchromatic, and the pan-sharpened image bands for a best color representation. a statistical approached was applied to the pan-sharpening process for standardizing and automating the pan-sharpening process. the new algorithm is available within the pci geomatics software. usually the panchromatic and multispectral images have to be processed separately to ensure both data are aligned with each other before pansharpening. these steps include reading of data, collection of ground controls (gcps), and orthorectification with a dem. since the gcps are collected separately and the images are orthorectified separately, it may cause a slight misalignment between the orthorectified panchromatic and multispectral images due to the gcp location and distribution. one advantage of the ikonos geo-ortho kit product is that the panchromatic and multispectral images are resampled exactly on top of each other. in general, it is possible to perform pan-sharpening of the data first for gentle terrain before further processing. thus, the user would only need to perform gcp collection and orthorectification once to the pan-sharpened image. to test this method, pan-sharpening was first applied to the geo-ortho kit data. figure 1a shows the urban area example of the latest news? visit www.geoinformatics.com multispectral data at 4m resolution, figure 1b shows the corresponding panchromatic image at 1m resolution, and figure 1c shows the pan-sharpened image. figures 2a, 2b and 2c show similar examples for a semiurban area. both pan-sharpened results did not show any problems and verified that it is possible to pan-sharpen the geo-ortho kit data first. test results a total of 79 gcps and 40 tie points were collected using the 1:10000 vectors provided. according to the united states national map accuracy standards, the accuracy of 1:10000 vector is approximately 8.5m ce90 (approximately 4m rms). ideally, the accuracy of gcps should be within sub-meter accuracy. the elevations for the gcps were extracted from the dem provided. additionally, as previously mentioned, the dem was also used by the tie points to change the points from planimetric to altimetric points during the bundle adjustment to improve accuracy. several cases were used to test the accuracy and distribution of gcps. first all gcps were changed to independent check points (icps). then the following cases were tested by changing some of the icps into gcps: (1) no gcps, (2) one gcp per image, (3) two gcps per image, (4) three gcps per image, (5) 1 gcp per corner image of the block, (6) two gcps per corner image of the block, (7) one gcp per bottom corner image of the block, and (8) two gcps per bottom corner image of the block. table 1 shows a summary of the results. . the table illustrates how the accuracy is affected by the number and distribution of gcps. in general, it is better to collect at least one or two gcps per image. collecting more gcps per image did not improve the result significantly. an alternative is to collect at least one gcp per corner image of the block when the number of gcps is limited. both cases would produce results within 4m rms, i.e. within the accuracy of the gcps. collecting gcps on just one side of the block is not recommended for high accuracy. march 2008 55
Powered by fluidbook tm